gate-58 e2e-networkidle matches COMMENT lines — larpingapp reports 1 finding with zero live calls

[rubenvdlinde]

Same shape as gate-48 (#191): the checker greps for the pattern with no comment filter, so a comment warning against networkidle is reported as a use of it.

Measured

Full-tree run of run-hydra-gates.sh (hydra-gates main @ 756fe89) against ConductionNL/larpingapp@development c73df92d:

[gate-58] e2e-networkidle: FAIL — 1 networkidle wait(s) in changed e2e file(s)

The single log line is:

tests/e2e/spec-coverage/detail-forms-admin.spec.ts:187:	// live `waitForLoadState('networkidle')` in the suite; every other mention

In context (lines 185-189):

// ADR-074 rule 4: `networkidle` never settles on Nextcloud — the
// notification poll keeps the network permanently busy. This was the LAST
// live `waitForLoadState('networkidle')` in the suite; every other mention
// is a comment warning against it.

The repo has zero live calls. Every one of the 10 occurrences under tests/ is a comment:

$ grep -rn "networkidle" tests/ src/ | grep -vE "^\S+:[0-9]+:\s*(//|\*|/\*|#)"
$ echo $?
1        # no matches

Cause

run-hydra-gates.sh around line 4515:

grep -nE "waitForLoadState\([[:space:]]*['\"]networkidle['\"]|waitUntil:[[:space:]]*['\"]networkidle['\"]" "$f" \
    | grep -v "e2e-networkidle exclude" \
    | while IFS= read -r hit; do ...

The only filter is the exclude marker. There is no comment stripping, so // … waitForLoadState('networkidle') … matches exactly like a call.

Note this is a doubly perverse failure mode: the more carefully a repo documents why it removed its networkidle waits, the more findings it accrues. larpingapp is the repo the gate was written against (its index-pages.spec.ts is cited in the gate header), so it has the most such comments in the fleet.

Suggested fix

Strip line comments before matching — drop lines whose match position sits after a //, *, or # on the same line. A minimal version that would fix this case without touching the pattern:

grep -nE "<pattern>" "$f" \
    | grep -v "e2e-networkidle exclude" \
    | grep -vE "^[0-9]+:[[:space:]]*(//|\*|/\*)" \

That still catches a real call with a trailing comment, and stops counting prose. A code-aware strip would be better but this closes the observed shape.

Impact

Any repo that documented the ADR-074 rule 4 removal in a comment now carries a permanent, unfixable gate-58 finding — unfixable because the only way to clear it is to delete the explanation of why the code is correct, or to add an exclude marker to a comment, which is comment-satisfaction.

[rubenvdlinde]

Fixed and merged to main in #269 (cdfbd7a). Live fleet-wide now.

Measured post-merge on larpingapp, attribution-isolated between main@32fa23e (immediately before the merge) and main@cdfbd7a, read from the runner's stdout:

A (before): [gate-58] e2e-networkidle: FAIL — 1 networkidle wait(s) in changed e2e file(s)
B (after):  [gate-58] e2e-networkidle: PASS

Zero changes in larpingapp. Your line is a unit fixture now, verbatim.

On the minimal fix you sketched (grep -vE "^[0-9]+:[[:space:]]*(//|\*|/\*)") — deliberately not taken, and there are tests for why:

• it still counts const x = 1 // waitForLoadState('networkidle') is banned — a mention after code on the same line;
• it still counts a block-comment interior line that begins with a letter, e.g. waitForLoadState('networkidle') must never be used. between /* and */;
• it loses a real call carrying a trailing comment.

The mask blanks the comment regions, whatever column they start in, and preserves offsets — which is what makes the e2e-networkidle exclude marker still work, since that marker lives in a comment, exactly what the mask removes. It is read from the original line at the matched line number. Asserted both ways, including that an exclude on a neighbouring line does not suppress.

openbuild in the same run: 22 → 14. All 8 removed were verified individually as comments explaining why networkidle is not used; the 14 that remain are live calls. That is the doubly-perverse effect you named — the better a repo documented the ADR-074 rule 4 removal, the redder it got.